When closing timing on challenging designs, users may choose to run many
different place_design directives in order to select the best timing result. Auto
directives use machine learning to predict the best directives to run. Users can benefit
by only running these directives instead of the full sweep of directives listed in
Available
Directives.
Note: When running with the Auto
directives, the directive setup happens slightly later in the flow than when the
directive is directly specified which can result is slightly different
results.
Machine learning prediction of directives have a margin of error. As a
consequence, it is recommended to run 3 Auto_n directives and take
the best result. The directives predicted are equivalent to the ones mentioned in
Available
Directives, so there is no benefit in running auto directives in
addition to these. The directive selected by the tool is reported in the log file. An
example of the message is as
follows:
INFO: [Place 30-1947] Predicted directive using ML models is: EarlyBlockPlacementTo enable the feature, set the place_design -directive
<value> where value is:
- Auto_1
- High performing predicted directive
- Auto_2
- Second best predicted directive
- Auto_3
- Third best predicted directive